KirschBluteX/engineer-software
A runtime-neutral, evidence-driven software-engineering workflow for AI coding agents: for substantive work it routes a request to exactly one of six bounded workflows and defines the evidence required before the agent may change direction or claim completion — Shape Work (contract and exclusions when behaviour/scope is unclear), Trace Failure (reproduction plus causal evidence), Probe Choice (a disposable experiment and its decision consequence), Deliver Change (focused check and final-state evidence), Inspect Structure (traced owners/callers plus a boundary recommendation) and Manage Work Items (a local PRD/task set with acceptance criteria). Ordinary explanations, translations, formatting changes and already-specified mechanical operations are explicitly bypassed. Codex and DeepSeek Harness load the same canonical SKILL.md, references and routing cases; the Harness projection is generated and checked from the Codex source rather than maintained twice.
Install
python scripts/sync_harness_skill.py --write && npx @deepseek-ai/dsh webREADME gives two runtime entries into one canonical workflow. DeepSeek Harness: the checkout includes a generated .dsh/skills/engineer-software projection — check it with python scripts/sync_harness_skill.py --check and python scripts/validate_harness.py --check, start npx @deepseek-ai/dsh web, then choose this repository as the workspace; after a git pull or an edit to the canonical skill, regenerate with --write and re-validate. A user-global copy can target $DSH_HOME/skills/engineer-software, and removal is Remove-Item -LiteralPath .dsh/skills/engineer-software -Recurse after confirming the target. Codex: codex plugin marketplace add KirschBluteX/engineer-software, codex plugin add engineer-software@engineer-software, then start a new task. The README states it is not an official DeepSeek plugin, partnership or endorsement and does not invent a harness manifest outside the documented filesystem skill contract.
Compatibility
DeepSeek Harness (developer preview; the README warns compatibility-breaking changes are possible) and/or Codex. Python 3.9+ with requirements-dev.txt for the validators. The .dsh/skills tree is a generated projection of the Codex canonical source — drift fails validation.
Details
- Repo: KirschBluteX/engineer-software
- Category: Other
- Stars: 6
- Version: Skill/workflow repository (no npm package). Validation uses Python 3.9+.
- Last push: 2026-08-15
- First seen: 2026-08-07
Recent updates
The README documents its validation surface: deterministic routing checks without model access (python scripts/validate_evals.py, validate_harness.py --check, run_routing_eval.py --limit 5), optional read-only live Codex evidence, and an aggregate validate_project.py covering the plugin package, routing fixtures, Harness projection and documentation contracts. It states plainly that task-level A/B runs are sampled behavioural evidence, not a benchmark claim, and that the README publishes no single speedup percentage.
FAQ
- How do I install Engineer Software for DeepSeek Harness?
- The checkout ships a generated .dsh/skills/engineer-software projection. Verify it with python scripts/sync_harness_skill.py --check and python scripts/validate_harness.py --check, start the harness, then open this repository as the workspace. Regenerate with --write after updating the canonical skill.
- How does it decide which workflow to use?
- A router first checks whether the request is ordinary work or has material engineering uncertainty; substantive work starts exactly one of six bounded workflows (Shape Work, Trace Failure, Probe Choice, Deliver Change, Inspect Structure, Manage Work Items), each with explicit evidence required to leave it. Plain explanations and mechanical edits bypass the workflow.
- Is this an official DeepSeek plugin?
- No. The README states it is not an official DeepSeek plugin, partnership or endorsement, and that it does not ship an MCP server, hook, telemetry, credential store or background service; tool permissions, API keys and model configuration stay under your runtime policy.